In Plato’s Phaedrus, Socrates relates a conversation between the Egyptian god Theuth, who invented writing, and the mythical King Thamus.
Theuth claimed that writing would improve wisdom. King Thamus disagreed:
For this invention will produce forgetfulness in the minds of those who learn to use it, because they will not practice their memory. Their trust in writing, produced by external characters which are no part of themselves, will discourage the use of their own memory within them. You have invented an elixir not of memory, but of reminding; and you offer your pupils the appearance of wisdom, not true wisdom, for they will read many things without instruction and will therefore seem to know many things, when they are for the most part ignorant and hard to get along with, since they are not wise, but only appear wise.
Simply put: this new-fangled writing will lead to brainrot.
What side are we supposed to take in this debate — should we be for, or against, writing?
Socrates is our touchstone, and his view was about the same as the king’s:
He who thinks, then, that he has left behind him any art in writing, and he who receives it in the belief that anything in writing will be clear and certain, would be an utterly simple person, and in truth ignorant of the prophecy of Ammon, if he thinks written words are of any use except to remind him who knows the matter about which they are written.
Again, to put it simply: Socrates says we can’t learn anything new from reading. We can only be reminded of what we already know.
This notion seems all wrong to me. Granted, I’m only reading a text whose contents I have no formal instruction in, and so I’m precisely the sort of ignoramus they’re debating. But the premise is seemingly that knowledge is obtainable only from a live teacher, not a textual one. Otherwise, the objection that a text can only remind us of what we already know, and not yield knowledge itself, wouldn’t make sense. As someone who learns best from solitary reading, I find this hard to relate to.
What they must mean is that only reading a text is insufficient to fully understand it. A text by itself, this argument would go, is static. Its assertions must be tested, argued, questioned, defended, and applied before we truly absorb them. I still think this overstates; while reading, we perform many of those processes implicitly. But there’s something to it. In a sense, it’s a defense of Socrates’s own method of discovery through dialogue. At the very least, taking something we’ve read in a book and applying it to a different context — whether that’s through verbal debate, or in our own writing — unlocks deeper understanding.
When I encountered those passages from Phaedrus, essentially arguing that writing will make us stupider because we’ll skip over the work of true understanding, I of course thought of AI.
What is the worry there? One is that AI will make us stupider because we’re no longer doing the hard work of thinking. We’re having something else us think for us. Like a muscle that weakens with disuse, our brains will atrophy.
Yet, do we think human intellect went downhill from Plato’s time, when memory recall was replaced with resort to written texts? No, I don’t think anyone could seriously argue that. You might argue that our memory faculty has weakened since then, but other faculties developed: our ability to synthesize and put to creative use the larger bodies of knowledge available from texts, for one. Through writing, and its corollary, reading, human thought was no longer constrained by the amount of information held in a single mind.
On the other hand, AI use is fundamentally different from the turn toward writing. Now we’re outsourcing not the storage of knowledge, but thinking itself — that is, the use of knowledge to derive ideas, opinions, and conclusions.
The effect is not a weaker memory (well, maybe that too). It’s weaker judgment, weaker analytical capacity.
I’m still not sure despair is in order. Writing, when it came, unlocked new kinds of intellectual skill and production. It’s entirely possible that AI will do something similar: weaken certain faculties, yes, but also enable new ones.
What persuades me most is that, as far into the past as we have visibility, humans have devoted vast time and energy to thinking. Not only to resolve the problems that need resolving (like how to keep warm or preserve food), but to find answers to questions that don’t immediately aid our survival (like who created the earth and the cosmos; and why is Frank so annoying). In fact, we regularly create problems in order to give ourselves things to think about. (How else do you explain most of Substack?) For many of us, we’re not content unless we’re analyzing, investigating, figuring, opining. We won’t stop using our minds in challenging ways now that we have AI. It’s a compulsion; we’re powerless to resist it.
Early research into the effect of AI on our thinking abilities suggests that how we use AI is key. That is, whether we use it as a thinking substitute, or a thinking partner.
My best thinking often emerges in dialogue. Explaining my thoughts to someone else is how my best insights arise.1 As well, when voicing my ideas to someone else, I can hear them as if at a distance, and better critically assess them.
(This is why simply reading your work aloud to someone else aids self-editing. You hear your words as a reader would and detect awkward phrasing, redundant sentences, and explanatory gaps. Somehow, it transforms your angle of attention — this also accounts for the annoying phenomenon where typos remain hidden until after you send the email, and reread it as sent.2)
Photos by Hert Niks, Dawid Tkocz, & The Cleveland Museum of Art on Unsplash
Thought that emerges through dialogue: this takes us back to the ancients. The Socratic dialogues perform thought through conversation — thought emerges as an entity outside of a single person’s mind, as something suspended in the air, made of the exhalations of multiple speakers. A new faculty enabled through AI is the ability to dialogue with a smart, well-informed partner at any moment. We don’t have to be alone with our thoughts, which tend to stagnate locked inside the mind’s cave. We can air them out, expose them, test them, by speaking them to another who can lob them back at a different velocity and angle. We can take what we’ve learned from a text and apply it in discourse, coincidentally overcoming Socrates’s objection to the one-dimensionality of reading.
I use AI daily in my work as a tech lawyer. We’re required to, but I also enjoy it. AI is a dead-end for original analysis, but it’s highly generative as a thought partner. I’m not playing tennis with myself. There’s someone — something — sending volleys of thought back, forcing me to shift my stance and angle of engagement.
Given my general optimism that humans will continue to put their brains to productive (and non-productive) use no matter what, my worries about AI are different. I don’t think our brains will rot — or at least, not those of us inclined to exercise them regularly. (As for the rest, their brains are probably already rotting, and I’m not sure AI will make a significant difference. But I tend to think those people are few and far between, and I think it’s good that people are designed with different appetites for intellectualizing. We don’t need more takes on Yesteryear or What Would The Hunter-Gatherers Do; the tank’s full!3)
Here’s my worry. Imagine if, 500 years after the shift from memory stores to textual stores that Phaedrus worried over, writing technology suddenly vanished. (Truly a hypothetical, because I have no idea how you un-discover writing.) Without knowledge stored in their minds, and without the mnemonic techniques that the ancients practiced when the only accessible library was in one’s head, humans would have been immensely set back.
Once we habitually outsource certain baseline thinking to AI, we shift to a new faculty of thinking that’s dependent on it. It’s easier to imagine AI technology disappearing of a sudden than writing. What then? Potentially, we’d lose access to the kind of thinking we would have come to depend on, and, like the mnemonic techniques of the ancients lost through disuse, we’d be unable to fall back immediately to the more basic form of thinking that was replaced by AI. An alarming thought. But still, we’d figure it out — eventually.
I’m curious — if you use AI, how do you use it? I know some readers abstain completely. But I haven’t heard much beyond that. I’m particularly interested in your take on the thought “substitution” versus thought “partnership” framing.
Today I’ve written about our compulsion to think. Last fall, I wrote about a different kind of compulsion: to arrest time through visual records.
Back when I was starting out on Substack, I wrote a short piece about AI. Returning to it now, what hits me: my writing and thinking have improved a lot since April 2025!
There’s a connection here to Charlie Finch’s recent piece on why AI writing feels dead. His theory is that human writing is emergent; one line gives rise to another. What’s written is not predetermined, it’s discovered in-process. Whereas AI writing is predetermined, inorganic. Similar to the way thought emerges through writing, it also emerges through discoursing. I suspect that the neurological processes underlying both overlap; each practice involves translating thought into words intended for an audience.
I learned the read-aloud technique from Susan Bell’s The artful edit: On the practice of editing yourself (2008). Speaking of AI, here’s a trick that builds on her advice: Prompt ChatGPT to repeat — exactly — your own draft back to you. Then use the option to listen to ChatGPT’s response; you’ll hear your own words read aloud. Like false notes in a familiar song, many problems will reveal themselves — ones that you passed over multiple times during silent revision.
I would like to hear some solutions to the demographic crisis that don’t involve twisting people’s arms to have more children. If anyone’s looking for assignment!
I never thought about it this way. I guess with all new technologies, it's a form of empowerment. Some people use it to help, and others use it to hurt people. The invention of the printing press allowed Malleus Maleficarum to be easily shared, leading to far more witch hunts. However, printing presses meant knowledge could finally disseminate in a real way. I guess it was hard for Socrates to imagine a positive use for writing because he hadn't seen one, yet. He seemed to be coming at it from a place of privilege, imagining that anyone who wanted to learn through discourse had access to a thinking partner. Now, here we are, we've finally figured out how to give more people those thinking partners, which maybe makes reading less important, but I have to say, I enjoyed reading your post. What would Socrates say, I wonder?
Thanks for the article shout out!
Also, I use AI, probably much in the same way you do. For legal work it is incredibly helpful iterating arguments, cutting through large tranches of data and documents, and testing theories and arguments for holes. And, unlike creative works, I feel you can cross reference various legal theories and get it to come to conclusions it might not have developed on its own. Or at the very least to give you a foothold on where the theory might gain purchase or be slippery. The development of AI trained specifically on statutes and case law is extremely helpful in this regard, as general purpose AI tends to overapply laws and be very general in its thinking.
I can also see how people may use it to dialogue about ideas. I don't really use it that way, but I can see how it may help someone free up ideas by iterating with it. I am concerned about that type of use and how it impacts the users, though. There are a lot of emotionally and socially vulnerable people who are already easy subjects of manipulation from less advanced sources (ads, politics, news bias, etc.). AI creeps along with no motive other than to execute its function, which is to please and generate more interaction.
It operates on a layer of manipulation people aren't trained to see because there is no obvious intent. Many people get changed by the AI (believe the sycophancy) and then begin to overvalue what in the end are often average ideas simply because of the confidence of how they are delivered and the belief that the tech is truly smart in its judgment, rather than smarty in predictive ability.